In modern agriculture, the harvesting of fruits and vegetables remains a labor-intensive process, accounting for approximately 40% of the total workload in crop production. Automation through robotic systems offers a promising solution, but a critical challenge lies in designing end effectors that can handle delicate produce like tomatoes without causing damage. This article presents the development and testing of a fully pneumatic, suction-gripping integrated end effector for tomato harvesting. The design focuses on achieving stable clamping and non-destructive picking through innovative mechanical structure and intelligent control strategies. As a researcher in agricultural robotics, I have been involved in the design, simulation, and field validation of this end effector, aiming to address the non-structured environments and variability in fruit characteristics that complicate mechanical harvesting.
The core of this end effector is a spatial multi-link three-finger mechanism driven by a single pneumatic cylinder. This design integrates two sequential actions—fruit suction-retraction and finger clamping—into one actuation cycle, simplifying the system and enhancing reliability. The end effector consists of three identical gripper units symmetrically arranged around a central axis, each comprising linkages and sliders that convert the linear motion of the pneumatic cylinder into the coordinated opening and closing of the fingers. A vacuum cup is attached to a central slider, which moves axially as the fingers close, pulling the fruit inward to ensure proper alignment and reduce positional errors. This integrated approach minimizes the number of active components, which is beneficial for robustness and cost-effectiveness in field applications.
To optimize the geometry of the end effector, a kinematic model was established. The motion equations for key points, such as the suction point K and the finger tip H, were derived based on geometric relationships. For instance, the position of the suction point K is given by:
$$y_K = h + \sqrt{L^2 – (x_C – x_A)^2} – \sqrt{l_1^2 – (x_C – x_A – l_6)^2}, \quad x_C \in [l, l+c]$$
where \(y_K\) is the vertical coordinate, \(h\) is the distance from point A to K, \(L\) and \(l_1\) are lengths of the lower and upper linkages, \(x_C\) is the horizontal coordinate of the slider, \(x_A\) is the coordinate of point A, \(l_6\) is a fixed length, \(l\) is the initial position, and \(c\) is the slider travel distance. Similarly, the coordinates of the finger hinge point D and tip H can be expressed as functions of \(x_C\). These equations allow for the determination of structural parameters that satisfy operational requirements, such as a maximum finger envelope diameter of 156 mm and a maximum suction cup retraction distance of 38.7 mm. The key parameters are summarized in Table 1.
| Parameter | Value (mm) | Parameter | Value (mm or degrees) |
|---|---|---|---|
| l | 40 | l_7 | 10 |
| l_1 | 30 | x_A | 10 |
| l_2 | 60 | L | 80 |
| l_3 | 11.7 | c | 14 |
| l_4 | 25 | h | 40 |
| l_5 | 53 | L_{max} | 156 |
| l_6 | 35.5 | L_{min} | 0.5 |
| θ_3 | 90° | θ_4 | 129° |
| θ_5 | 152° | h_{max} | 38.7 |
Following the kinematic analysis, dynamic simulations were conducted using ADAMS software to evaluate the motion and force characteristics of the end effector. A spherical model with tomato-like properties—such as a longitudinal stiffness of \(1.9198 \times 10^3 \, \text{N/m}\), density of \(1.04 \, \text{g/cm}^3\), elastic modulus of \(1.7866 \times 10^6 \, \text{N/m}^2\), and Poisson’s ratio of 0.14—was used to represent the fruit. The simulation model included constraints, drives, and contact forces between the fingers and the fruit. Results showed that during clamping, the lower linkage exhibited minimal velocity and acceleration changes, while the upper linkage and suction cup had significant movements in the later stages. The relationship between clamping force, system pressure, and fruit size was also analyzed. For a 100 mm diameter fruit, the clamping force \(F_c\) varies linearly with pneumatic pressure \(P\):
$$F_c = k \cdot P$$
where \(k\) is a constant dependent on the mechanism geometry. At 0.6 MPa, the clamping force reached approximately 44 N, while at 0.1 MPa, it was about 8.8 N. Additionally, for a fixed pressure of 0.4 MPa, the clamping force increased with fruit size due to greater contact area. These insights highlight the need for adaptive pressure control to accommodate varying fruit dimensions and avoid damage.
The pneumatic system for the end effector was designed with feedback control to enable real-time adjustment of clamping force. It includes a 5 L air reservoir, filters, regulators, and valves to supply three separate circuits: one for the rotary cylinder (for fruit twisting), one for the three-finger cylinder (for clamping), and one for the vacuum generator (for suction). An Arduino Due controller manages the system, using a proportional pressure valve (SMC ITV2050-312L) to regulate the clamping pressure continuously from 0.005 to 0.9 MPa. Force-sensitive resistors (FSR-402) embedded in each finger provide real-time pressure feedback, allowing for closed-loop control. This setup ensures that the end effector can respond dynamically to prevent slippage and minimize mechanical stress on the fruit.
Central to the non-destructive operation is a slip detection algorithm based on pressure sensor signals. Experiments were conducted where tomatoes were manually pulled or rotated while clamped at low pressure (0.1 MPa). The pressure data from the sensors were analyzed in the time domain. During slippage, the sensor readings exhibited sudden fluctuations and declines. To reliably detect these events, a dual-threshold algorithm was developed using the dynamic standard deviation (SD) of the sensor data. The dynamic SD at data point \(x\) is calculated as:
$$\text{SD}(x) = \sqrt{\frac{\sum_{i=1}^{x} (n(i) – \text{Dyavg}(x))^2}{x}}, \quad x \leq \text{lengthdata}$$
where \(n(i)\) is the sensor reading array, and \(\text{Dyavg}(x)\) is the dynamic average. A slip event is identified when the increment in SD exceeds a threshold (set to 0.1) and persists for a minimum duration (set to 6 consecutive samples). This method reduces false triggers compared to single-threshold approaches. Based on this slip criterion, a non-destructive clamping control strategy was implemented: starting at a low pressure of 0.1 MPa, the system pressure increases by 0.05 MPa each time slippage is detected, up to a limit of 0.6 MPa (determined from static damage tests to prevent bruising). This iterative process ensures minimal clamping force while maintaining a secure grip.
Field trials were conducted to evaluate the performance of the end effector. A total of 204 ripe tomatoes of the Hongyun 721 variety were harvested, categorized into four size groups: 80–100 mm, 70–80 mm, 60–70 mm, and 50–60 mm in diameter. The end effector was mounted on an AUBO-i5 robotic arm equipped with a stereo vision camera for fruit localization. Each picking cycle involved suction, retraction, clamping, and twisting actions, controlled by the Arduino system. Success rates, direct damage rates (immediate mechanical injury), and 72-hour browning rates (indicative of latent damage) were recorded. The results are summarized in Table 2.
| Size Range (mm) | Number of Tomatoes | Success Rate (%) | Direct Damage Rate (%) | 72-Hour Browning Rate (%) |
|---|---|---|---|---|
| 80–100 | 31 | 90.32 | 3.22 | 0.00 |
| 70–80 | 55 | 96.36 | 1.82 | 1.89 |
| 60–70 | 78 | 97.44 | 1.28 | 2.63 |
| 50–60 | 40 | 100.00 | 0.00 | 2.50 |
| Overall Average | 204 | 96.03 | 1.58 | 1.76 |
The overall success rate of 96.03% demonstrates the end effector’s effectiveness in harvesting tomatoes across different sizes. The average picking time per fruit was 5 seconds, which is competitive for robotic systems. Direct damage was primarily observed in larger fruits, often due to incomplete suction contact or interference from stems and leaves, leading to unstable clamping. Browning rates were low, suggesting that the control strategy successfully limited excessive force. These outcomes validate the design and control approach, showing that the end effector can operate reliably in real-world conditions while preserving fruit quality.
In conclusion, this work presents a pneumatic end effector for tomato picking that integrates suction and clamping into a single mechanism driven by one actuator. The kinematic design ensures adequate envelope range and retraction distance, while dynamic simulations provide insights into force transmission. The electro-pneumatic control system, coupled with a slip-based algorithm, enables adaptive clamping force adjustment to prevent damage. Field tests confirm high success rates and low damage incidence, meeting practical requirements for automated harvesting. Future improvements could focus on enhancing suction reliability for irregular fruit shapes and optimizing the control parameters for other produce types. This end effector represents a step forward in agricultural robotics, offering a robust and non-destructive solution for fruit harvesting.